Abstract:

Vowel perception is influenced by the fundamental frequency of voicing (F0 normalization). This paper describes a similar effect resulting from a manipulation of phonation type. Acoustic characteristics of two phonation types were produced in formant-synthesized and LPC resynthesized stimuli. In the formant-synthesized stimuli, spectral tilt and open quotient were used to produce a flat-spectrum continuum (low-amplitude H1 and high-amplitude spectrum above 1000 Hz) and a tilted-spectrum continuum (high-amplitude H1 and low-amplitude higher frequencies). The flat-spectrum tokens sound like a pressed or creaky voice while the tilted-spectrum tokens sound breathy. In the LPC resynthesized stimuli a similar effect was achieved by replacing the LPC residual with voices synthesized using the LF model of voicing. Identification responses to both sets of stimuli showed reliable boundary shifts on an F1 ``hood''--``hud'' continuum. There were more ``hood'' responses to the flat-spectrum continua than to the tilted-spectrum continua (higher F1 boundaries). This voice quality effect is reminiscent of F0 speaker normalization, but the creaky voice (which sounded more like a male speaker) was associated with a higher F1 boundary. Data on perceived speaker characteristics of the stimuli will be reported. [Work supported by NIDCD R29-DC01645.]